txindex.go

// Copyright (c) 2016 The btcsuite developers
// Copyright (c) 2016-2019 The Decred developers
// Use of this source code is governed by an ISC
// license that can be found in the LICENSE file.

package indexers

import (
	"context"
	"errors"
	"fmt"

	"github.com/decred/dcrd/chaincfg/chainhash"
	"github.com/decred/dcrd/database/v2"
	"github.com/decred/dcrd/dcrutil/v3"
	"github.com/decred/dcrd/wire"
)

const (
	// txIndexName is the human-readable name for the index.
	txIndexName = "transaction index"

	// txIndexVersion is the current version of the transaction index.
	txIndexVersion = 2

	// txEntrySize is the size of a transaction entry.  It consists of 4
	// bytes block id + 4 bytes offset + 4 bytes length + 4 bytes block
	// index.
	txEntrySize = 4 + 4 + 4 + 4
)

var (
	// txIndexKey is the key of the transaction index and the db bucket used
	// to house it.
	txIndexKey = []byte("txbyhashidx")

	// idByHashIndexBucketName is the name of the db bucket used to house
	// the block id -> block hash index.
	idByHashIndexBucketName = []byte("idbyhashidx")

	// hashByIDIndexBucketName is the name of the db bucket used to house
	// the block hash -> block id index.
	hashByIDIndexBucketName = []byte("hashbyididx")

	// errNoBlockIDEntry is an error that indicates a requested entry does
	// not exist in the block ID index.
	errNoBlockIDEntry = errors.New("no entry in the block ID index")
)

// -----------------------------------------------------------------------------
// The transaction index consists of an entry for every transaction in the main
// chain.  In order to significantly optimize the space requirements a separate
// index which provides an internal mapping between each block that has been
// indexed and a unique ID for use within the hash to location mappings.  The ID
// is simply a sequentially incremented uint32.  This is useful because it is
// only 4 bytes versus 32 bytes hashes and thus saves a ton of space in the
// index.
//
// There are three buckets used in total.  The first bucket maps the hash of
// each transaction to the specific block location.  The second bucket maps the
// hash of each block to the unique ID and the third maps that ID back to the
// block hash.
//
// NOTE: Although it is technically possible for multiple transactions to have
// the same hash as long as the previous transaction with the same hash is fully
// spent, this code only stores the most recent one because doing otherwise
// would add a non-trivial amount of space and overhead for something that will
// realistically never happen per the probability and even if it did, the old
// one must be fully spent and so the most likely transaction a caller would
// want for a given hash is the most recent one anyways.
//
// The serialized format for keys and values in the block hash to ID bucket is:
//   <hash> = <ID>
//
//   Field           Type              Size
//   hash            chainhash.Hash    32 bytes
//   ID              uint32            4 bytes
//   -----
//   Total: 36 bytes
//
// The serialized format for keys and values in the ID to block hash bucket is:
//   <ID> = <hash>
//
//   Field           Type              Size
//   ID              uint32            4 bytes
//   hash            chainhash.Hash    32 bytes
//   -----
//   Total: 36 bytes
//
// The serialized format for the keys and values in the tx index bucket is:
//
//   <txhash> = <block id><start offset><tx length><block index>
//
//   Field           Type              Size
//   txhash          chainhash.Hash    32 bytes
//   block id        uint32            4 bytes
//   start offset    uint32            4 bytes
//   tx length       uint32            4 bytes
//   block index     uint32            4 bytes
//   -----
//   Total: 48 bytes
// -----------------------------------------------------------------------------

// TxIndexEntry houses information about an entry in the transaction index.
type TxIndexEntry struct {
	// BlockRegion specifies the location of the raw bytes of the transaction.
	BlockRegion database.BlockRegion

	// BlockIndex species the index of the transaction within the array of
	// transactions that comprise a tree of the block.
	BlockIndex uint32
}

// dbPutBlockIDIndexEntry uses an existing database transaction to update or add
// the index entries for the hash to id and id to hash mappings for the provided
// values.
func dbPutBlockIDIndexEntry(dbTx database.Tx, hash *chainhash.Hash, id uint32) error {
	// Serialize the height for use in the index entries.
	var serializedID [4]byte
	byteOrder.PutUint32(serializedID[:], id)

	// Add the block hash to ID mapping to the index.
	meta := dbTx.Metadata()
	hashIndex := meta.Bucket(idByHashIndexBucketName)
	if err := hashIndex.Put(hash[:], serializedID[:]); err != nil {
		return err
	}

	// Add the block ID to hash mapping to the index.
	idIndex := meta.Bucket(hashByIDIndexBucketName)
	return idIndex.Put(serializedID[:], hash[:])
}

// dbRemoveBlockIDIndexEntry uses an existing database transaction remove index
// entries from the hash to id and id to hash mappings for the provided hash.
func dbRemoveBlockIDIndexEntry(dbTx database.Tx, hash *chainhash.Hash) error {
	// Remove the block hash to ID mapping.
	meta := dbTx.Metadata()
	hashIndex := meta.Bucket(idByHashIndexBucketName)
	serializedID := hashIndex.Get(hash[:])
	if serializedID == nil {
		return nil
	}
	if err := hashIndex.Delete(hash[:]); err != nil {
		return err
	}

	// Remove the block ID to hash mapping.
	idIndex := meta.Bucket(hashByIDIndexBucketName)
	return idIndex.Delete(serializedID)
}

// dbFetchBlockIDByHash uses an existing database transaction to retrieve the
// block id for the provided hash from the index.
func dbFetchBlockIDByHash(dbTx database.Tx, hash *chainhash.Hash) (uint32, error) {
	hashIndex := dbTx.Metadata().Bucket(idByHashIndexBucketName)
	serializedID := hashIndex.Get(hash[:])
	if serializedID == nil {
		return 0, errNoBlockIDEntry
	}

	return byteOrder.Uint32(serializedID), nil
}

// dbFetchBlockHashBySerializedID uses an existing database transaction to
// retrieve the hash for the provided serialized block id from the index.
func dbFetchBlockHashBySerializedID(dbTx database.Tx, serializedID []byte) (*chainhash.Hash, error) {
	idIndex := dbTx.Metadata().Bucket(hashByIDIndexBucketName)
	hashBytes := idIndex.Get(serializedID)
	if hashBytes == nil {
		return nil, errNoBlockIDEntry
	}

	var hash chainhash.Hash
	copy(hash[:], hashBytes)
	return &hash, nil
}

// dbFetchBlockHashByID uses an existing database transaction to retrieve the
// hash for the provided block id from the index.
func dbFetchBlockHashByID(dbTx database.Tx, id uint32) (*chainhash.Hash, error) {
	var serializedID [4]byte
	byteOrder.PutUint32(serializedID[:], id)
	return dbFetchBlockHashBySerializedID(dbTx, serializedID[:])
}

// putTxIndexEntry serializes the provided values according to the format
// described about for a transaction index entry.  The target byte slice must
// be at least large enough to handle the number of bytes defined by the
// txEntrySize constant or it will panic.
func putTxIndexEntry(target []byte, blockID uint32, txLoc wire.TxLoc, blockIndex uint32) {
	byteOrder.PutUint32(target, blockID)
	byteOrder.PutUint32(target[4:], uint32(txLoc.TxStart))
	byteOrder.PutUint32(target[8:], uint32(txLoc.TxLen))
	byteOrder.PutUint32(target[12:], blockIndex)
}

// dbPutTxIndexEntry uses an existing database transaction to update the
// transaction index given the provided serialized data that is expected to have
// been serialized putTxIndexEntry.
func dbPutTxIndexEntry(dbTx database.Tx, txHash *chainhash.Hash, serializedData []byte) error {
	txIndex := dbTx.Metadata().Bucket(txIndexKey)
	return txIndex.Put(txHash[:], serializedData)
}

// dbFetchTxIndexEntry uses an existing database transaction to fetch the block
// region for the provided transaction hash from the transaction index.  When
// there is no entry for the provided hash, nil will be returned for the both
// the region and the error.
func dbFetchTxIndexEntry(dbTx database.Tx, txHash *chainhash.Hash) (*TxIndexEntry, error) {
	// Load the record from the database and return now if it doesn't exist.
	txIndex := dbTx.Metadata().Bucket(txIndexKey)
	serializedData := txIndex.Get(txHash[:])
	if len(serializedData) == 0 {
		return nil, nil
	}

	// Ensure the serialized data has enough bytes to properly deserialize.
	if len(serializedData) < txEntrySize {
		return nil, database.Error{
			ErrorCode: database.ErrCorruption,
			Description: fmt.Sprintf("corrupt transaction index "+
				"entry for %s", txHash),
		}
	}

	// Load the block hash associated with the block ID.
	hash, err := dbFetchBlockHashBySerializedID(dbTx, serializedData[0:4])
	if err != nil {
		return nil, database.Error{
			ErrorCode: database.ErrCorruption,
			Description: fmt.Sprintf("corrupt transaction index "+
				"entry for %s: %v", txHash, err),
		}
	}

	// Deserialize the final entry.
	entry := TxIndexEntry{
		BlockRegion: database.BlockRegion{
			Hash:   new(chainhash.Hash),
			Offset: byteOrder.Uint32(serializedData[4:8]),
			Len:    byteOrder.Uint32(serializedData[8:12]),
		},
		BlockIndex: byteOrder.Uint32(serializedData[12:16]),
	}
	copy(entry.BlockRegion.Hash[:], hash[:])
	return &entry, nil
}

// dbAddTxIndexEntries uses an existing database transaction to add a
// transaction index entry for every transaction in the parent of the passed
// block (if they were valid) and every stake transaction in the passed block.
func dbAddTxIndexEntries(dbTx database.Tx, block *dcrutil.Block, blockID uint32) error {
	// The offset and length of the transactions within the serialized block.
	txLocs, stakeTxLocs, err := block.TxLoc()
	if err != nil {
		return err
	}

	// As an optimization, allocate a single slice big enough to hold all
	// of the serialized transaction index entries for the block and
	// serialize them directly into the slice.  Then, pass the appropriate
	// subslice to the database to be written.  This approach significantly
	// cuts down on the number of required allocations.
	addEntries := func(txns []*dcrutil.Tx, txLocs []wire.TxLoc, blockID uint32) error {
		offset := 0
		serializedValues := make([]byte, len(txns)*txEntrySize)
		for i, tx := range txns {
			putTxIndexEntry(serializedValues[offset:], blockID, txLocs[i],
				uint32(i))
			endOffset := offset + txEntrySize
			err := dbPutTxIndexEntry(dbTx, tx.Hash(),
				serializedValues[offset:endOffset:endOffset])
			if err != nil {
				return err
			}
			offset += txEntrySize
		}
		return nil
	}

	// Add the regular tree transactions.
	err = addEntries(block.Transactions(), txLocs, blockID)
	if err != nil {
		return err
	}

	// Add the stake tree transactions.
	return addEntries(block.STransactions(), stakeTxLocs, blockID)
}

// dbRemoveTxIndexEntry uses an existing database transaction to remove the most
// recent transaction index entry for the given hash.
func dbRemoveTxIndexEntry(dbTx database.Tx, txHash *chainhash.Hash) error {
	txIndex := dbTx.Metadata().Bucket(txIndexKey)
	serializedData := txIndex.Get(txHash[:])
	if len(serializedData) == 0 {
		return fmt.Errorf("can't remove non-existent transaction %s "+
			"from the transaction index", txHash)
	}

	return txIndex.Delete(txHash[:])
}

// dbRemoveTxIndexEntries uses an existing database transaction to remove the
// latest transaction entry for every transaction in the parent of the passed
// block (if they were valid) and every stake transaction in the passed block.
func dbRemoveTxIndexEntries(dbTx database.Tx, block *dcrutil.Block) error {
	removeEntries := func(txns []*dcrutil.Tx) error {
		for _, tx := range txns {
			err := dbRemoveTxIndexEntry(dbTx, tx.Hash())
			if err != nil {
				return err
			}
		}
		return nil
	}

	// Remove the regular and stake tree transactions from the block being
	// disconnected.
	if err := removeEntries(block.Transactions()); err != nil {
		return err
	}
	return removeEntries(block.STransactions())
}

// TxIndex implements a transaction by hash index.  That is to say, it supports
// querying all transactions by their hash.
type TxIndex struct {
	db         database.DB
	curBlockID uint32
}

// Ensure the TxIndex type implements the Indexer interface.
var _ Indexer = (*TxIndex)(nil)

// Init initializes the hash-based transaction index.  In particular, it finds
// the highest used block ID and stores it for later use when connecting or
// disconnecting blocks.
//
// This is part of the Indexer interface.
func (idx *TxIndex) Init() error {
	// Find the latest known block id field for the internal block id
	// index and initialize it.  This is done because it's a lot more
	// efficient to do a single search at initialize time than it is to
	// write another value to the database on every update.
	err := idx.db.View(func(dbTx database.Tx) error {
		// Scan forward in large gaps to find a block id that doesn't
		// exist yet to serve as an upper bound for the binary search
		// below.
		var highestKnown, nextUnknown uint32
		testBlockID := uint32(1)
		increment := uint32(100000)
		for {
			_, err := dbFetchBlockHashByID(dbTx, testBlockID)
			if err != nil {
				nextUnknown = testBlockID
				break
			}

			highestKnown = testBlockID
			testBlockID += increment
		}
		log.Tracef("Forward scan (highest known %d, next unknown %d)",
			highestKnown, nextUnknown)

		// No used block IDs due to new database.
		if nextUnknown == 1 {
			return nil
		}

		// Use a binary search to find the final highest used block id.
		// This will take at most ceil(log_2(increment)) attempts.
		for {
			testBlockID = (highestKnown + nextUnknown) / 2
			_, err := dbFetchBlockHashByID(dbTx, testBlockID)
			if err != nil {
				nextUnknown = testBlockID
			} else {
				highestKnown = testBlockID
			}
			log.Tracef("Binary scan (highest known %d, next "+
				"unknown %d)", highestKnown, nextUnknown)
			if highestKnown+1 == nextUnknown {
				break
			}
		}

		idx.curBlockID = highestKnown
		return nil
	})
	if err != nil {
		return err
	}

	log.Debugf("Current internal block ID: %d", idx.curBlockID)
	return nil
}

// Key returns the database key to use for the index as a byte slice.
//
// This is part of the Indexer interface.
func (idx *TxIndex) Key() []byte {
	return txIndexKey
}

// Name returns the human-readable name of the index.
//
// This is part of the Indexer interface.
func (idx *TxIndex) Name() string {
	return txIndexName
}

// Version returns the current version of the index.
//
// This is part of the Indexer interface.
func (idx *TxIndex) Version() uint32 {
	return txIndexVersion
}

// Create is invoked when the indexer manager determines the index needs
// to be created for the first time.  It creates the buckets for the hash-based
// transaction index and the internal block ID indexes.
//
// This is part of the Indexer interface.
func (idx *TxIndex) Create(dbTx database.Tx) error {
	meta := dbTx.Metadata()
	if _, err := meta.CreateBucket(idByHashIndexBucketName); err != nil {
		return err
	}
	if _, err := meta.CreateBucket(hashByIDIndexBucketName); err != nil {
		return err
	}
	_, err := meta.CreateBucket(txIndexKey)
	return err
}

// ConnectBlock is invoked by the index manager when a new block has been
// connected to the main chain.  This indexer adds a hash-to-transaction mapping
// for every transaction in the passed block.
//
// This is part of the Indexer interface.
func (idx *TxIndex) ConnectBlock(dbTx database.Tx, block, parent *dcrutil.Block, _ PrevScripter, isTreasuryEnabled bool) error {
	// NOTE: The fact that the block can disapprove the regular tree of the
	// previous block is ignored for this index because even though the
	// disapproved transactions no longer apply spend semantics, they still
	// exist within the block and thus have to be processed before the next
	// block disapproves them.
	//
	// Also, the transaction index is keyed by hash and only supports a single
	// transaction per hash.  This means that if the disapproved transaction
	// is mined into a later block, as is typically the case, only that most
	// recent one can be queried.  Ideally, it should probably support multiple
	// transactions per hash, which would not only allow access in the case
	// just described, but it would also allow indexing of transactions that
	// happen to have the same hash (granted the probability of this is
	// extremely low), which is supported so long as the previous one is
	// fully spent.

	// Increment the internal block ID to use for the block being connected
	// and add all of the transactions in the block to the index.
	newBlockID := idx.curBlockID + 1
	if err := dbAddTxIndexEntries(dbTx, block, newBlockID); err != nil {
		return err
	}

	// Add the new block ID index entry for the block being connected and
	// update the current internal block ID accordingly.
	err := dbPutBlockIDIndexEntry(dbTx, block.Hash(), newBlockID)
	if err != nil {
		return err
	}
	idx.curBlockID = newBlockID
	return nil
}

// DisconnectBlock is invoked by the index manager when a block has been
// disconnected from the main chain.  This indexer removes the
// hash-to-transaction mapping for every transaction in the block.
//
// This is part of the Indexer interface.
func (idx *TxIndex) DisconnectBlock(dbTx database.Tx, block, parent *dcrutil.Block, _ PrevScripter, isTreasuryEnabled bool) error {
	// NOTE: The fact that the block can disapprove the regular tree of the
	// previous block is ignored when disconnecting blocks because it is also
	// ignored when connecting the block.  See the comments in ConnectBlock for
	// the specifics.

	// Remove all of the transactions in the block from the index.
	if err := dbRemoveTxIndexEntries(dbTx, block); err != nil {
		return err
	}

	// Remove the block ID index entry for the block being disconnected and
	// decrement the current internal block ID to account for it.
	if err := dbRemoveBlockIDIndexEntry(dbTx, block.Hash()); err != nil {
		return err
	}
	idx.curBlockID--
	return nil
}

// Entry returns details for the provided transaction hash from the transaction
// index.  The block region contained in the result can in turn be used to load
// the raw transaction bytes.  When there is no entry for the provided hash, nil
// will be returned for the both the entry and the error.
//
// This function is safe for concurrent access.
func (idx *TxIndex) Entry(hash *chainhash.Hash) (*TxIndexEntry, error) {
	var entry *TxIndexEntry
	err := idx.db.View(func(dbTx database.Tx) error {
		var err error
		entry, err = dbFetchTxIndexEntry(dbTx, hash)
		return err
	})
	return entry, err
}

// NewTxIndex returns a new instance of an indexer that is used to create a
// mapping of the hashes of all transactions in the blockchain to the respective
// block, location within the block, and size of the transaction.
//
// It implements the Indexer interface which plugs into the IndexManager that in
// turn is used by the blockchain package.  This allows the index to be
// seamlessly maintained along with the chain.
func NewTxIndex(db database.DB) *TxIndex {
	return &TxIndex{db: db}
}

// dropBlockIDIndex drops the internal block id index.
func dropBlockIDIndex(db database.DB) error {
	return db.Update(func(dbTx database.Tx) error {
		meta := dbTx.Metadata()
		err := meta.DeleteBucket(idByHashIndexBucketName)
		if err != nil {
			return err
		}

		return meta.DeleteBucket(hashByIDIndexBucketName)
	})
}

// DropTxIndex drops the transaction index from the provided database if it
// exists.  Since the address index relies on it, the address index will also be
// dropped when it exists.
func DropTxIndex(ctx context.Context, db database.DB) error {
	// Nothing to do if the index doesn't already exist.
	exists, err := existsIndex(db, txIndexKey, txIndexName)
	if err != nil {
		return err
	}
	if !exists {
		log.Infof("Not dropping %s because it does not exist", txIndexName)
		return nil
	}

	// Mark that the index is in the process of being dropped so that it
	// can be resumed on the next start if interrupted before the process is
	// complete.
	err = markIndexDeletion(db, txIndexKey)
	if err != nil {
		return err
	}

	// Drop the address index if it exists, as it depends on the transaction
	// index.
	err = DropAddrIndex(ctx, db)
	if err != nil {
		return err
	}

	log.Infof("Dropping all %s entries.  This might take a while...",
		txIndexName)

	// Since the indexes can be so large, attempting to simply delete
	// the bucket in a single database transaction would result in massive
	// memory usage and likely crash many systems due to ulimits.  In order
	// to avoid this, use a cursor to delete a maximum number of entries out
	// of the bucket at a time.
	err = incrementalFlatDrop(ctx, db, txIndexKey, txIndexName)
	if err != nil {
		return err
	}

	// Call extra index specific deinitialization for the transaction index.
	err = dropBlockIDIndex(db)
	if err != nil {
		return err
	}

	// Remove the index tip, version, bucket, and in-progress drop flag now
	// that all index entries have been removed.
	err = dropIndexMetadata(db, txIndexKey, txIndexName)
	if err != nil {
		return err
	}

	log.Infof("Dropped %s", txIndexName)
	return nil
}

// DropIndex drops the transaction index from the provided database if it
// exists.  Since the address index relies on it, the address index will also be
// dropped when it exists.
func (*TxIndex) DropIndex(ctx context.Context, db database.DB) error {
	return DropTxIndex(ctx, db)
}